linux-kernel - Re: [PATCH v2 2/2] can: spi: hi311x: Add Holt HI-311x CAN driver

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Thu, 9 Mar 2017 10:28:36 -0500
From:   Akshay Bhat <akshay.bhat@...esys.com>
To:     Wolfgang Grandegger <wg@...ndegger.com>, mkl@...gutronix.de
Cc:     linux-can@...r.kernel.org, netdev@...r.kernel.org,
        devicetree@...r.kernel.org, linux-kernel@...r.kernel.org,
        Akshay Bhat <nodeax@...il.com>
Subject: Re: [PATCH v2 2/2] can: spi: hi311x: Add Holt HI-311x CAN driver

Hi Wolfgang,

On 03/09/2017 09:45 AM, Wolfgang Grandegger wrote:
> Hello Akshay,
> 
> Am 09.03.2017 um 13:34 schrieb Akshay Bhat:
>>
>> Hi Wolfgang,
>>
>> Good question. I have not worked with MPC251x but the HI-311x performs
>> much better because HI-3110 features:
>> 8 message FIFO (as opposed to 2 buffers on MPC2510)
>> 20 MHz SPI interface (as opposed to 2.5 MHz on MPC2510)
>>
>> As for the real world test results:
>>
>> With RT patch applied to the kernel running on a i.MX6 Dual processor
>> (worst case interrupt latency of 50us as reported by cyclictest), there
>> are ZERO packet drops.
>> Tested with Kvaser Leaf sending 100 burst messages (back to back) every
>> 40ms at a 1M CAN bit rate. 10 million messages were sent by the Kvaser
>> leaf and received successfully by the HI-311x driver.
> 
> This corresponds to a bus load of approx. 50%, I think?

The usecase I tested above was more like ~30% (data payload length was
randomized) but I did test with 50% bus load as well without any dropped
messages.

> 
>> Even without the RT patch, I was able to get the packet drop to zero but
>> this was by moving the CAN/SPI IRQ threads to CPU1 instead of CPU0.
> 
> Vanilla Linux is more critical here due to higher latencies. With 2500
> Messages per sec the RX FIFO (8 Messages) fills up within 3.2 ms... and
> in a burst even quicker. That's already heavy load.
> 

I agree with Vanilla Linux is more critical due to higher latencies,
however if the frequency governor is set to performance and CAN/SPI
threads are moved to another CPU (if that is an option) with SCHED_FIFO
setting, the driver is able to keep up even at high bit rates / bursts.

I understand SPI based CAN controllers are not ideal choice but for
users that do not have any other option, I feel this controller/driver
meets their need :)

Thanks,
Akshay